外观
LangChain RAG 实践指南
LangChain 是一个强大的框架,用于构建基于大语言模型的应用。本指南将详细介绍如何使用 LangChain 构建 RAG(检索增强生成)应用。
概述
什么是 LangChain?
LangChain 是一个开源框架,提供了构建 LLM 应用的工具和组件:
- 组件化设计:模块化的组件便于组合
- 链式调用:支持复杂的调用链
- 记忆管理:内置对话记忆功能
- 工具集成:丰富的工具和集成
- RAG 支持:完整的 RAG 实现方案
核心组件
- Document Loaders: 文档加载器
- Text Splitters: 文本切分器
- Embeddings: 向量化模型
- Vector Stores: 向量存储
- Retrievers: 检索器
- Chains: 调用链
- Memory: 记忆管理
环境准备
安装依赖
bash
pip install langchain langchain-openai langchain-community
pip install chromadb faiss-cpu
pip install pypdf unstructured配置 API Key
python
import os
os.environ["OPENAI_API_KEY"] = "your-api-key"快速开始
基础 RAG 实现
1. 加载文档
python
from langchain.document_loaders import PyPDFLoader
from langchain.document_loaders import TextLoader
from langchain.document_loaders import DirectoryLoader
def load_documents():
loader = PyPDFLoader("document.pdf")
documents = loader.load()
return documents
def load_directory(directory_path):
loader = DirectoryLoader(
directory_path,
glob="**/*.pdf",
loader_cls=PyPDFLoader
)
documents = loader.load()
return documents2. 文本切分
python
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain.text_splitter import CharacterTextSplitter
def split_documents(documents):
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000,
chunk_overlap=200,
length_function=len,
separators=["\n\n", "\n", " ", ""]
)
chunks = text_splitter.split_documents(documents)
return chunks3. 创建向量存储
python
from langchain_openai import OpenAIEmbeddings
from langchain_community.vectorstores import Chroma
def create_vector_store(documents):
embeddings = OpenAIEmbeddings()
vector_store = Chroma.from_documents(
documents=documents,
embedding=embeddings,
persist_directory="./chroma_db"
)
return vector_store4. 创建检索器
python
def create_retriever(vector_store):
retriever = vector_store.as_retriever(
search_type="similarity",
search_kwargs={"k": 5}
)
return retriever5. 创建问答链
python
from langchain_openai import ChatOpenAI
from langchain.chains import RetrievalQA
def create_qa_chain(retriever):
llm = ChatOpenAI(
model_name="gpt-3.5-turbo",
temperature=0.7
)
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=retriever,
return_source_documents=True
)
return qa_chain6. 完整示例
python
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_community.vectorstores import Chroma
from langchain.chains import RetrievalQA
def build_rag_system(file_path):
loader = PyPDFLoader(file_path)
documents = loader.load()
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000,
chunk_overlap=200
)
chunks = text_splitter.split_documents(documents)
embeddings = OpenAIEmbeddings()
vector_store = Chroma.from_documents(
documents=chunks,
embedding=embeddings
)
retriever = vector_store.as_retriever(
search_kwargs={"k": 5}
)
llm = ChatOpenAI(temperature=0.7)
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=retriever,
return_source_documents=True
)
return qa_chain
qa_chain = build_rag_system("document.pdf")
result = qa_chain({"query": "文档的主要内容是什么?"})
print(result["result"])高级功能
自定义 Prompt
python
from langchain.prompts import PromptTemplate
def create_custom_prompt():
prompt_template = """
基于以下上下文信息回答用户问题。如果上下文中没有相关信息,请明确说明"根据提供的信息无法回答该问题"。
上下文信息:
{context}
用户问题:{question}
请提供详细、准确的回答:
"""
PROMPT = PromptTemplate(
template=prompt_template,
input_variables=["context", "question"]
)
return PROMPT
def create_custom_qa_chain(retriever, prompt):
llm = ChatOpenAI(temperature=0.7)
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=retriever,
chain_type_kwargs={"prompt": prompt},
return_source_documents=True
)
return qa_chain多轮对话
python
from langchain.memory import ConversationBufferMemory
from langchain.chains import ConversationalRetrievalChain
def create_conversational_rag(vector_store):
llm = ChatOpenAI(temperature=0.7)
memory = ConversationBufferMemory(
memory_key="chat_history",
return_messages=True,
output_key="answer"
)
qa_chain = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=vector_store.as_retriever(),
memory=memory,
return_source_documents=True,
verbose=True
)
return qa_chain
qa_chain = create_conversational_rag(vector_store)
chat_history = []
while True:
query = input("用户: ")
if query.lower() in ["exit", "quit"]:
break
result = qa_chain({"question": query, "chat_history": chat_history})
print(f"AI: {result['answer']}")
chat_history.append((query, result['answer']))混合检索
python
from langchain.retrievers import EnsembleRetriever
from langchain.retrievers import BM25Retriever
def create_hybrid_retriever(documents, vector_store):
bm25_retriever = BM25Retriever.from_documents(documents)
bm25_retriever.k = 5
vector_retriever = vector_store.as_retriever(
search_kwargs={"k": 5}
)
ensemble_retriever = EnsembleRetriever(
retrievers=[bm25_retriever, vector_retriever],
weights=[0.4, 0.6]
)
return ensemble_retriever重排序
python
from langchain.retrievers import ContextualCompressionRetriever
from langchain.retrievers.document_compressors import LLMChainExtractor
def create_reranking_retriever(base_retriever):
llm = ChatOpenAI(temperature=0)
compressor = LLMChainExtractor.from_llm(llm)
compression_retriever = ContextualCompressionRetriever(
base_compressor=compressor,
base_retriever=base_retriever
)
return compression_retriever流式输出
python
from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler
def create_streaming_qa_chain(retriever):
llm = ChatOpenAI(
temperature=0.7,
streaming=True,
callbacks=[StreamingStdOutCallbackHandler()]
)
qa_chain = RetrievalQA.from_chain_type(
llm=llm,
chain_type="stuff",
retriever=retriever
)
return qa_chain向量数据库集成
Chroma
python
from langchain_community.vectorstores import Chroma
vector_store = Chroma.from_documents(
documents=chunks,
embedding=embeddings,
persist_directory="./chroma_db"
)
vector_store.persist()FAISS
python
from langchain_community.vectorstores import FAISS
vector_store = FAISS.from_documents(
documents=chunks,
embedding=embeddings
)
vector_store.save_local("./faiss_index")
vector_store = FAISS.load_local(
"./faiss_index",
embeddings
)Milvus
python
from langchain_community.vectorstores import Milvus
vector_store = Milvus.from_documents(
documents=chunks,
embedding=embeddings,
collection_name="langchain_demo",
connection_args={
"host": "localhost",
"port": "19530"
}
)Pinecone
python
from langchain_community.vectorstores import Pinecone
import pinecone
pinecone.init(
api_key="your-api-key",
environment="your-environment"
)
vector_store = Pinecone.from_documents(
documents=chunks,
embedding=embeddings,
index_name="langchain-index"
)文档加载器
PDF 文档
python
from langchain.document_loaders import PyPDFLoader
loader = PyPDFLoader("document.pdf")
documents = loader.load()Word 文档
python
from langchain.document_loaders import Docx2txtLoader
loader = Docx2txtLoader("document.docx")
documents = loader.load()网页内容
python
from langchain.document_loaders import WebBaseLoader
loader = WebBaseLoader("https://example.com")
documents = loader.load()CSV 文件
python
from langchain.document_loaders import CSVLoader
loader = CSVLoader("data.csv")
documents = loader.load()JSON 文件
python
from langchain.document_loaders import JSONLoader
loader = JSONLoader(
file_path="data.json",
jq_schema=".messages[]",
content_key="content"
)
documents = loader.load()文本切分策略
递归字符切分
python
from langchain.text_splitter import RecursiveCharacterTextSplitter
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000,
chunk_overlap=200,
length_function=len,
separators=["\n\n", "\n", " ", ""]
)字符切分
python
from langchain.text_splitter import CharacterTextSplitter
text_splitter = CharacterTextSplitter(
separator="\n\n",
chunk_size=1000,
chunk_overlap=200
)代码切分
python
from langchain.text_splitter import PythonCodeTextSplitter
text_splitter = PythonCodeTextSplitter(
chunk_size=1000,
chunk_overlap=200
)Markdown 切分
python
from langchain.text_splitter import MarkdownTextSplitter
text_splitter = MarkdownTextSplitter(
chunk_size=1000,
chunk_overlap=200
)性能优化
缓存机制
python
from langchain.cache import InMemoryCache
from langchain.globals import set_llm_cache
set_llm_cache(InMemoryCache())
llm = ChatOpenAI(temperature=0)批处理
python
def batch_process_documents(documents, batch_size=100):
from langchain_community.vectorstores import Chroma
vector_store = None
for i in range(0, len(documents), batch_size):
batch = documents[i:i + batch_size]
if vector_store is None:
vector_store = Chroma.from_documents(
documents=batch,
embedding=embeddings
)
else:
vector_store.add_documents(batch)
return vector_store异步处理
python
import asyncio
from langchain_community.vectorstores import Chroma
async def async_add_documents(vector_store, documents):
await vector_store.aadd_documents(documents)
async def async_search(vector_store, query):
results = await vector_store.asimilarity_search(query, k=5)
return results实战案例
企业知识库问答
python
from langchain.document_loaders import DirectoryLoader
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain_community.vectorstores import Chroma
from langchain.chains import ConversationalRetrievalChain
from langchain.memory import ConversationBufferMemory
class EnterpriseKnowledgeBase:
def __init__(self, documents_dir):
self.documents_dir = documents_dir
self.qa_chain = None
def load_documents(self):
loader = DirectoryLoader(
self.documents_dir,
glob="**/*.pdf",
loader_cls=PyPDFLoader
)
return loader.load()
def build_index(self):
documents = self.load_documents()
text_splitter = RecursiveCharacterTextSplitter(
chunk_size=1000,
chunk_overlap=200
)
chunks = text_splitter.split_documents(documents)
embeddings = OpenAIEmbeddings()
self.vector_store = Chroma.from_documents(
documents=chunks,
embedding=embeddings,
persist_directory="./knowledge_base"
)
def create_qa_system(self):
llm = ChatOpenAI(temperature=0.7)
memory = ConversationBufferMemory(
memory_key="chat_history",
return_messages=True
)
self.qa_chain = ConversationalRetrievalChain.from_llm(
llm=llm,
retriever=self.vector_store.as_retriever(),
memory=memory,
return_source_documents=True
)
def query(self, question):
result = self.qa_chain({"question": question})
return {
"answer": result["answer"],
"sources": result["source_documents"]
}
kb = EnterpriseKnowledgeBase("./documents")
kb.build_index()
kb.create_qa_system()
result = kb.query("公司的年假政策是什么?")
print(result["answer"])常见问题
1. 内存占用过高
问题: 处理大量文档时内存不足
解决方案:
- 使用批处理
- 启用持久化存储
- 减小 chunk_size
2. 检索质量不佳
问题: 检索到的文档不相关
解决方案:
- 调整切分策略
- 使用混合检索
- 添加重排序
3. 响应速度慢
问题: 系统响应延迟高
解决方案:
- 启用缓存
- 使用异步处理
- 优化向量数据库
最佳实践
1. 文档处理
- 选择合适的文档加载器
- 优化切分参数
- 保留文档元数据
2. 检索优化
- 使用混合检索策略
- 配置合理的检索参数
- 实现重排序机制
3. 用户体验
- 支持多轮对话
- 提供来源引用
- 实现流式输出
参考资料
官方文档
相关教程
提示
LangChain 提供了丰富的组件和灵活的组合方式,建议根据具体需求选择合适的组件。
信息
本指南基于 LangChain 最新版本编写,API 可能随版本更新而变化,请参考官方文档。
